wav2vec2-common_voice-tr-demo-dist

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the COMMON_VOICE - TR dataset. It achieves the following results on the evaluation set:

• Loss: 0.3934
• Wer: 0.3305

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 8
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.5459	0.23	100	3.6773	1.0
3.2247	0.46	200	3.1491	0.9999
2.3457	0.69	300	2.4236	1.0041
0.9149	0.92	400	0.9471	0.7684
0.6622	1.15	500	0.7518	0.6863
0.7205	1.38	600	0.6387	0.6402
0.6978	1.61	700	0.5611	0.5739
0.5317	1.84	800	0.5061	0.5418
0.5222	2.07	900	0.4839	0.5344
0.4467	2.3	1000	0.5060	0.5339
0.3196	2.53	1100	0.4619	0.5213
0.276	2.76	1200	0.4595	0.5020
0.3569	2.99	1300	0.4339	0.4901
0.2236	3.22	1400	0.4602	0.4887
0.293	3.45	1500	0.4376	0.4639
0.1677	3.68	1600	0.4371	0.4605
0.1838	3.91	1700	0.4116	0.4589
0.1225	4.14	1800	0.4144	0.4495
0.2301	4.37	1900	0.4250	0.4567
0.1931	4.6	2000	0.4081	0.4470
0.1427	4.83	2100	0.4295	0.4482
0.361	5.06	2200	0.4374	0.4445
0.3272	5.29	2300	0.4088	0.4258
0.3686	5.52	2400	0.4087	0.4258
0.3087	5.75	2500	0.4100	0.4371
0.4637	5.98	2600	0.4038	0.4219
0.1485	6.21	2700	0.4361	0.4197
0.1341	6.44	2800	0.4217	0.4132
0.1185	6.67	2900	0.4244	0.4097
0.1588	6.9	3000	0.4212	0.4181
0.0697	7.13	3100	0.3981	0.4073
0.0491	7.36	3200	0.3992	0.4010
0.088	7.59	3300	0.4206	0.4022
0.0731	7.82	3400	0.3998	0.3841
0.2767	8.05	3500	0.4195	0.3829
0.1725	8.28	3600	0.4167	0.3946
0.1242	8.51	3700	0.4177	0.3821
0.1133	8.74	3800	0.3993	0.3802
0.1952	8.97	3900	0.4132	0.3904
0.1399	9.2	4000	0.4010	0.3795
0.047	9.43	4100	0.4128	0.3703
0.049	9.66	4200	0.4319	0.3670
0.0994	9.89	4300	0.4118	0.3631
0.1209	10.11	4400	0.4296	0.3722
0.0484	10.34	4500	0.4130	0.3615
0.2065	10.57	4600	0.3958	0.3668
0.133	10.8	4700	0.4102	0.3679
0.0622	11.03	4800	0.4137	0.3585
0.0999	11.26	4900	0.4042	0.3583
0.0346	11.49	5000	0.4183	0.3573
0.072	11.72	5100	0.4060	0.3530
0.0365	11.95	5200	0.3968	0.3483
0.0615	12.18	5300	0.3958	0.3485
0.1067	12.41	5400	0.3987	0.3453
0.0253	12.64	5500	0.4182	0.3405
0.0636	12.87	5600	0.4199	0.3458
0.0506	13.1	5700	0.4056	0.3412
0.0944	13.33	5800	0.4061	0.3381
0.1187	13.56	5900	0.4113	0.3381
0.0237	13.79	6000	0.3973	0.3343
0.0166	14.02	6100	0.4001	0.3357
0.1189	14.25	6200	0.3931	0.3315
0.0375	14.48	6300	0.3944	0.3329
0.0537	14.71	6400	0.3953	0.3308
0.045	14.94	6500	0.3933	0.3303

Framework versions

• Transformers 4.18.0
• Pytorch 1.9.1+cu102
• Datasets 1.13.3
• Tokenizers 0.11.6